AgentSkillsCN

adk-grounding

精通ADK智能体在Python中与外部数据（Google搜索、Vertex AI搜索、RAG）的对接与融合。可用于实现可验证的响应、实时网页访问，以及私有知识的检索。

SKILL.md
--- frontmatter
name: adk-grounding
description: Expert in grounding ADK agents with external data (Google Search, Vertex AI Search, RAG) in Python. Use for implementing verifiable responses, real-time web access, and private knowledge retrieval.

ADK Grounding Specialist (Python Edition)

Philosophy & Architecture

Grounding connects agents to external sources to reduce hallucinations and provide citations.

Grounding Patterns

  1. •Google Search (Real-Time):
    • •Use GoogleSearchTool for news, weather, and up-to-the-minute facts.
  2. •Vertex AI Search (Enterprise):
    • •Use VertexAiSearchTool to query private PDFs, docs, and intranets.
    • •Requires a configured Data Store in Google Cloud.
  3. •Agentic RAG:
    • •Dynamic query construction and metadata filtering (e.g., Vector Search 2.0).

Implementation

  • •Configure grounding via the LlmAgent tools list.
  • •Ensure the agent is instructed to use grounding for specific types of queries.
  • •Read references/grounding.md for tool configuration details.

Success Criteria

  • •Valid citation metadata in agent responses.
  • •Correct configuration of Vertex AI Search data_store_id.
  • •High grounding accuracy as measured by hallucinations_v1 metrics.